Starting last month strategic polling firms have started to begin surveying the favorite presidential candidates of likely Iowa Caucus-goers. Being a non-resident of Iowa, I am very unfamiliar with how both the Presidential Caucus process actually works nationally, and how registered voters become eligible to participate in this major political event. However, I am more troubled by the early polls showing contradictory results not only at the national and state levels, but more so of surveys among likely Iowa caucus-goers.
For example, over the period of (December-January-February) four major polling firms conducted their first round of surveys among likely Iowa caucus-goers. The results show that John Edwards is in first place in three of the four polls (Strategic Vision, Zogby, Research 2000), as well as having a sizeable lead at both prospectively 8%, (Strategic Vision) and 10%, (Zogby) in two of these polls.
On the other hand, Edwards & Obama are tied for first in one poll (Research 2000), while (The American Research Group’s) is polling Hillary Clinton with a commanding eleven point lead in another poll over the second place candidate. (As a side note, American Research Group shows as of today Hillary’s lead in Iowa has grown from 11%, to 17%, over John Edwards.) The question now becomes what should any political junkie make of the polling when there is such a large difference in results.
POLLING RESULTS
Polls that have Edwards in First Place and Clinton in Fourth Place
Zogby (1/15-1/16) Strategic Vision (1/19-1/21) Research 2000 (12/18-12/20)
Zogby SV Reserach 2000 Averages
Edwards 27% Edwards 25% Edwards 22% 24% (+4.5)
Obama 17% Obama 17% Obama 22% 19.5% (-4.5)
Vilsack 16% Vilsack 16% Vilsack 12% 14% (-10)
Clinton 16% Clinton 15% Clinton 10% 13.67%(-10.33)
Poll that has Clinton in First Place and Edwards in Second Place
American Research Group (12/19-12/23)/ 1/31-2/1)
Clinton 31% Clinton 37%
Edwards 20% Edwards 18%
Vilsack 17% Obama 14%
Obama 10% Vilsack 12%
Trendline: Clinton + 6 points, Obama +4points, Edwards –2 points (HRC’s lead has grown from 11% to 19%, which is a 29.33% reversal from the average polling results of Zogby, Strategic Vision, and Research 2000.)
February 3, 2007-American Research Group Iowa Analysis
["While the race in Iowa is close among those saying they have attended at least one caucus (59% of likely caucus goers) with Clinton at 29% and Edwards at 25%, Clinton leads with 44% among those saying they will attend their first caucus in 2008 (Vilsack is at 17%, Obama is at 12%, and Edwards is at 7% with this group)."- American Reserach Group]
What factors are making The American Research Groups’ results so contradictory to the other polling firms?
Why are first time Iowa caucus-goers so highly favorable of Hillary Clinton as compared to those who are experienced ones?
Firstly, I asked myself what gives in the polling that allows for a wide variance in results at such an early point in the election. My first suggestion would normally be there must be a large difference in size of the margins of error between the polls. My analysis shows there isn’t a large enough variation in order to skew the results. In this case, each polling firms methodology uses a MOE that is about exactly +/- 4%. (Which is considered to be highly efficient and scientifically effective) As a result, from the sampling error shown between the polls being identical, the grey area for mathematical divergence is almost scientifically nullified. Therefore, I must be oversimplifying the accuracy of the polls or there must be other aspects of the polling process that are being highly ignored by me.
Secondly, when analyzing the different polling results what comes next to mind is what methodologies firms are using to prospectively identify likely voters. The methods typically used by prospective firms sampling the same political race are usually similar in how they all contact voters through screened telephone interviews, sample size of the electorate, and how they are chosen by random. Consequently, the differences in polling results among caucus-goers should normally on average be small and much of the variation comes from random sampling.
What is baffling in the case of 2007 Iowa caucus-goers, even with all four major polling firms using what seems to be similar previously stated polling tactics (telephone interviews, sample size,) and having the exactly same margin of error, there are still contradictory results among the different polls that are mathematically significant.
If anybody has info on how these specific firms create their voter sample please feel free to disclose that inform
Thirdly,
Technical Factors In Order of Importance
A.) "Turnout model"
-How pollsters create voter samples based on the voting age of the adult population through telephone interviews
Nonresponse voters (How pollsters tend to create sub-samples of the population based on expected turnout of registered adult voters)
Oversample voters based on statistically significant levels of non-response. (Voters who hang up or refuse to participate in political surveys are less likely to vote.
-Percentages of particular subgroups that are likely to vote (see down below)
B.) How live interviewers are framing their questions regarding which candidate voter’s favor.
(Limited political research has shown that firms that use pre-recorded voices rather than live humans generated more accurate results, but in the case the firms polling Iowa they all use real people.)
C.) Overall statistical weighing of subgroups
D.) Margin of Error
E.) Sample Size
F.) Voter reluctance
G.) Selection probabilities in proportion to population size within specific zip-codes/area-codes
H.) Voters misrepresenting which candidate they support
I.) Voters may be voting based on (name recognition/soft support) and are very willing to change the candidate they support
FACTORS THAT MAY BE IMPACTING POLLS
Sub-group factors and statistical weighting of these sub-sets
-Percentage of voters who are female
-Percentage of voters who are minority/white
-Percentage of voters above 25 years old
-Percentage of voters in different age groupings
Percentage of voters who are firsttime caucus goers
-Percentage of voters who have voted in a Caucus
-Percentage of union workers
-Percentage of college educated voters
-Percentage of Democrats/Independents/Republicans
In my opinion, political polling is an inexact science due to fact that no polling firm has a perfect method or technique on defining likely voters. As a result, pollsters truly lack a foolproof way to identify likely voters especially since when they in many cases exaggerate their tendency to vote, or purposely/non-purposely misrepresent their willingness to support one candidate over the other. Therefore, there is no mathematically solvent way to screen voters when no two voters are created equal.